[Linux-cluster] WebFarm using RedHat cluster suite ?

Fri Jan 20 16:12:15 UTC 2006

Thank you again for the input,
We will use FC 2TB SAN (after checking budget) with 10K FC HD.
Our Postgresql cluster will be active/passive (2 node connected to RAID
0+1 on one SAN group).
Servers inside the webfarm will be connected to RAID 5  on another SAN
group).

Marc Grimme wrote:

>On Thursday 19 January 2006 17:43, FM wrote:
>  
>
>>First, Thank you all for the great input !
>>Here is more output about our sites :
>>Some static HTML web site. Sites are generated in Lan and then rsync to
>>DMZ. And other with mod_perl + Postgresql Database (local)
>>    
>>
>Don't use postgres clustered(parallel with multiple writer nodes, HA should 
>not be a problem) on gfs. That is not supposed to work. As far as I know. You 
>would need a dbms that supports parallel clustering itself. Like i.e. Oracle 
>9iRAC.
>  
>
>>One of our most visited site (static html) is using nealy 200 GB of
>>bandwidth / month
>>and have 545368 hists a day.
>>For decembre 2005 stats  :
>>http://stats.lexum.umontreal.ca/awstats.pl?month=12&year=2005&output=main&c
>>onfig=www.canlii.org&lang=en&framename=index
>>
>>Now we are using dual Xeon (2 Ghz), with 2,5 GB of RAM. with RAID 5 SCSI
>>10KRPM
>>Network is GB
>>Budget is 200K CA$
>>    
>>
>Ok. You are having 500000 hit/day. One question would be how many I/Os does a 
>hit issue. The other question is, is that the upper bound or do you expect it 
>to grow, so that the infrastructure will have to grow in parallel. You should 
>really be picky about the storage-infrastructure (includes storage system and 
>locking network). Separate storage and locking network. Under normal 
>circumstances 500000hits/day should not be a problem. But if you are right 
>now using SCSI Disks stay with SCSI or FC-Disks they are way much faster and 
>relyable then SATA. The more and faster Disks you use the more I/Os you will 
>get. Besides for 200k CA$ you should also get a FC-Infrastructure and 
>Storage. 
>Also think about consolidating all data (even the os, share the root fs of the 
>servers) on to the storage system. It saves you loads of management (scales 
>within small amount of time) and disks in the servers. And you can easily 
>replace a faulty server by pulling it out and powering a new one back on. We 
>have made quite good experiences at webfarms with that concept. 
>Check back a http://www.open-sharedroot.org/. We will have a howto for 
>building up such a sharedroot within the next week.
>
>Hope that helps
>Regards Marc.
>  
>
>>Thanks again
>>
>>Eric Anderson wrote:
>>    
>>
>>>Marc Grimme wrote:
>>>      
>>>
>>>>Hello,
>>>>I think the best way to tell what storage or infrastructure would be
>>>>the best is to know more about your current setup and what issues
>>>>with that you want to get ahead of.
>>>>For example: if you really think about using iscsi, I don't think
>>>>that SCSI or SATA drives make a big difference - depending on how
>>>>many drives you use. But if all webservers currently have locally
>>>>attached disk drives you want scale too linar with exchanging and
>>>>IDE/parallel-SCSI Bus with an network topology using Ethernet. But my
>>>>opinion is: if you have a lot of I/Os make yourself mostly
>>>>independent from the latency of an ethernet and rethink about using
>>>>Fibre-Channel with GFS.
>>>>        
>>>>
>>>Honestly, ethernet latencies (especially on gigabit ethernet) are
>>>lower than fiber channel latencies, so this statement doesn't really
>>>hold up.
>>>If you want very fast speeds, get an iSCSI array, populate it with 15k
>>>RPM scsi disks with big caches, max the cache out on the array, and
>>>set it up for a RAID0+1 (or RAID10 depending on the implementor).  If
>>>you want fast speed, but not a big price, you can notch down to 10K
>>>scsi disks, or use 15k scsi disks and a raid 5, etc, and keep notching
>>>down until it fits your budget and needs.
>>>I agree here though that we really need to know a few things:
>>>
>>>- what kind of traffic is this?
>>>- size of the files most commonly used
>>>- total data size (how much space you need)
>>>- budget
>>>- demands (availability/performance/etc)
>>>
>>>Eric
>>>
>>>      
>>>
>>>>But the best advices could be made if you make your current setup and
>>>>the things you want to achieve more clearly.
>>>>
>>>>Regards Marc.
>>>>
>>>>On Wednesday 18 January 2006 22:01, FM wrote:
>>>>        
>>>>
>>>>>Thanks for the reply,
>>>>>I read about SATA storage but we sync from lan to dmz, so there is lots
>>>>>of r/w.
>>>>>
>>>>>Michael Will wrote:
>>>>>          
>>>>>
>>>>>>I am surprised you use SCSI drives on the storage if
>>>>>>you are price sensitive, usually SATA is the better
>>>>>>bang for the buck unless you are doing databases with
>>>>>>lots of small read and writes.
>>>>>>
>>>>>>Michael
>>>>>>
>>>>>>-----Original Message-----
>>>>>>From: linux-cluster-bounces at redhat.com
>>>>>>[mailto:linux-cluster-bounces at redhat.com] On Behalf Of FM
>>>>>>Sent: Wednesday, January 18, 2006 12:34 PM
>>>>>>To: Redhat Cluster
>>>>>>Subject: [Linux-cluster] WebFarm using RedHat cluster suite ?
>>>>>>
>>>>>>Hello everybody,
>>>>>>
>>>>>>Is redhat cluster suite (RHEL 4 ) a good candidate for a webfarm ?
>>>>>>My setup would be : several servers (1U AMd dual core) connected to
>>>>>>iscsi storage array.
>>>>>>
>>>>>>Is Iscsi a good choice (SAN prices are too high for us) for hardware ?
>>>>>>Our network is GB.
>>>>>>We will have SCSI 10KRPM + read and write cache on the SCSI card +
>>>>>>RAID5
>>>>>>thanks !
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>--
>>>>>>Linux-cluster mailing list
>>>>>>Linux-cluster at redhat.com
>>>>>>https://www.redhat.com/mailman/listinfo/linux-cluster
>>>>>>
>>>>>>--
>>>>>>Linux-cluster mailing list
>>>>>>Linux-cluster at redhat.com
>>>>>>https://www.redhat.com/mailman/listinfo/linux-cluster
>>>>>>            
>>>>>>
>>>>>--
>>>>>Linux-cluster mailing list
>>>>>Linux-cluster at redhat.com
>>>>>https://www.redhat.com/mailman/listinfo/linux-cluster
>>>>>          
>>>>>
>>--
>>Linux-cluster mailing list
>>Linux-cluster at redhat.com
>>https://www.redhat.com/mailman/listinfo/linux-cluster
>>    
>>
>
>  
>